Anyone can claim their current-affairs notes are "exam-relevant." We went back and checked — release by release, question by question, across every UPSC paper from 2021 to 2026 — and we're going to show you the numbers, including the ones that aren't flattering.
Most products quote one big number. We think that's where the dishonesty hides. A current-affairs filter has three separate jobs, and we score each on its own and publish them side by side. If we only showed you the friendliest one, you'd have every right to distrust the rest.
Did the source release make it into our edition before the exam? This is the selection job — finding the needle in 80,000.
Could a prepared aspirant answer the question from our card? Our headline metric — and the one that actually decides your marks.
Does the card alone decide all four options with zero outside knowledge? The hardest bar — and we'll tell you plainly where it caps out.
A backtest is worthless the moment you peek at the answer. It's easy to write a note after seeing the question and then declare it "100% accurate" — that proves nothing. So we built the test to make cheating impossible, the same way a serious research lab separates the people who build a model from the people who grade it.
Every card was written from the PIB release only — never from the question paper. The author literally could not see what the exam would ask.
A separate grader — who could not edit the card — scored it against the real question. The builder and the judge never talk.
The number we believe is the one from a year the method had never seen. Tuning on the past and testing on the future is the only honest test.
One honest thing up front, so every number below reads correctly. The UPSC Prelims paper is 100 questions, and most of it is static — history, polity, geography, economy and environment fundamentals — which no current-affairs product can or should claim to cover. We work only inside the current-affairs slice of the paper. And even there, a large share — global groupings, world sport and awards, geography-of-the-year — has no PIB source at all, so it's out of our scope by design. Saying no to that ~40% is part of the honesty.
A current-affairs source can only catch what it has seen. Early on, many tested schemes were founded before our archive even began — so the honest early numbers are thin, and they climb steeply as the corpus matures. This isn't a weakness we hide; it's the signature of a product that gets stronger every year.
On the most recent held-out paper, around 84% of the catchable questions were comfortably answerable from our card by a prepared aspirant. That's the number that maps to marks, and it's what we build for. The strict bar — card-decides-every-option with zero outside knowledge — sits near 45%, and there's a real structural ceiling there we will not pretend past.
Coaching wisdom says an exam draws on the last 12–16 months. Our data said otherwise — loudly. For the 2026 paper, the majority of the questions we caught came from releases older than 16 months; the oldest was a Cabinet decision from five years earlier. That single discovery is why our yearly booklet scans two windows instead of one — and why a competitor skimming only recent news would miss most of it.
A Mains question rarely "comes from" a press release. Instead PIB supplies the schemes, data, official problem-statements and way-forward you deploy inside a written answer. So we measure something different: of each year's questions, for how many did PIB carry usable material? We ran this across all four GS papers for four years — sixteen paper-years — verifying every cited source existed before that exam.
That coverage is really two kinds of help, and the difference is worth seeing. Some questions are an exact match — the question is literally about a scheme or body PIB covered (we call this direct). Most are surrounding material — PIB hands you the data, the example, the official problem and the way-forward you fold into a conceptual answer (referable enrichment). And some are genuinely out of scope — pure theory PIB has nothing to say about, which we count honestly as a miss, not a win.
And it held every single year — the ordering GS-III > GS-II > GS-I > GS-IV never once inverted across 2022–2025:
| 2022 | 2023 | 2024 | 2025 | mean | |
|---|---|---|---|---|---|
| GS-III | 90% | 100% | 95% | 95% | 95% |
| GS-II | 65% | 70% | 85% | 70% | 73% |
| GS-I | 35% | 45% | 45% | 70% | 49% |
| GS-IV | 24% | 37% | 26% | 26% | 28% |
A method that admits its limits is one you can trust on everything else. Here is where we stop.
Roughly 15–20% of tracked questions are structurally unwinnable from a single release — two-entity traps, exact trivia, match-the-pairs across unrelated items. We accept these rather than stuff cards with distractor bait.
We can't source a founding moment that predates our archive. Early editions are honestly weaker — and that's a maturing product, not a broken one.
Independent graders move borderline calls by a card or two each run. We trust the trend across years, never optimise to a single lucky grade.
You don't need the machinery to trust the result — but here's the honest shape of it. We read every English release. We score each on how strongly it names a thing an exam could test — a scheme, an Act, a body, an index, a mission — using both what the release looks like and what our system understands it to be. We select with guarantees so loud founding events never slip through. Then we enrich each card with the surrounding facts a complete note needs, and we fact-check every added number against at least two independent sources.
The exact signals, their weights, the entity-scoring rubric, and the rules we've distilled from five years of misses — that's the part that took the work, and it's the part we keep. We'll happily show you the results all day. The recipe stays ours.
🔒 Proprietary ranking model · refined after every real examThe fastest way to judge a filter is to read a day of it. A sample edition is free — no card written after seeing any question.
Read a sample edition →